diff --git a/benches/bench.rs b/benches/bench.rs
index d5915d4..b78b934 100644
--- a/benches/bench.rs
+++ b/benches/bench.rs
@@ -42,6 +42,18 @@ fn bench_usize_small(b: &mut Bencher) {
     });
 }
 
+#[bench]
+fn bench_to_bytes(b: &mut Bencher) {
+    let mut bit_vec = BitVec::from_elem(BENCH_BITS, false);
+    let mut r = small_rng();
+    for _ in 0..BENCH_BITS / 10 {
+        bit_vec.set((r.next_u32() as usize) % BENCH_BITS, true);
+    }
+    b.iter(|| {
+        black_box(bit_vec.to_bytes());
+    });
+}
+
 #[bench]
 fn bench_bit_set_big_fixed(b: &mut Bencher) {
     let mut r = small_rng();
diff --git a/src/lib.rs b/src/lib.rs
index ad32249..a3cc587 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1322,30 +1322,35 @@ impl<B: BitBlock> BitVec<B> {
     /// assert_eq!(bv.to_bytes(), [0b00100000, 0b10000000]);
     /// ```
     pub fn to_bytes(&self) -> Vec<u8> {
+        static REVERSE_TABLE: [u8; 256] = {
+            let mut tbl = [0u8; 256];
+            let mut i: u8 = 0;
+            loop {
+                tbl[i as usize] = i.reverse_bits();
+                if i == 255 {
+                    break;
+                }
+                i += 1;
+            }
+            tbl
+        };
         self.ensure_invariant();
-        // Oh lord, we're mapping this to bytes bit-by-bit!
-        fn bit<B: BitBlock>(bit_vec: &BitVec<B>, byte: usize, bit: usize) -> u8 {
-            let offset = byte * 8 + bit;
-            if offset >= bit_vec.nbits {
-                0
-            } else {
-                (bit_vec[offset] as u8) << (7 - bit)
+
+        let len = self.nbits / 8 + if self.nbits % 8 == 0 { 0 } else { 1 };
+        let mut result = Vec::with_capacity(len);
+
+        for byte_idx in 0..len {
+            let mut byte = 0u8;
+            for bit_idx in 0..8 {
+                let offset = byte_idx * 8 + bit_idx;
+                if offset < self.nbits && self[offset] {
+                    byte |= 1 << bit_idx;
+                }
             }
+            result.push(REVERSE_TABLE[byte as usize]);
         }
 
-        let len = self.nbits / 8 + if self.nbits % 8 == 0 { 0 } else { 1 };
-        (0..len)
-            .map(|i| {
-                bit(self, i, 0)
-                    | bit(self, i, 1)
-                    | bit(self, i, 2)
-                    | bit(self, i, 3)
-                    | bit(self, i, 4)
-                    | bit(self, i, 5)
-                    | bit(self, i, 6)
-                    | bit(self, i, 7)
-            })
-            .collect()
+        result
     }
 
     /// Compares a `BitVec` to a slice of `bool`s.